Geographical analysis and data mining
A characteristic feature in spatial analysis, contrary to normal data analysis, is that data sets have a location (2- or 3-dimensional), and that neighbourhood is important. Furthermore, spatial data (like nitrate concentration in the soil) may be sensitive to direction (anisotropic) due to slope of terrain, or direction of subsurface flow of water. Thirdly, there may be semantically governed restrictions between these class types (e.g., soil type A never occurs adjacent to soil type B, or land use X is always in the proximity of bodies of water). Another issue is that boundaries of geographic features or between two class types are often not crisp, and a transition zone may have to be incorporated in the modelling that precedes the analysis. Finally, the issue of scale is important: certain patterns or phenomena may only occur at a particular scale, and the derivation of this scale may be difficult in its own right. All of these issues influence algorithm design for geographical analysis and data mining (spatial and spatio-temporal data mining).

Data imprecision
Geographic data is always imprecise, and it is important to know in what respect the outcome of a geographic analysis depends on the imprecision. Algorithms that can determine this type of metadata are needed.

Industrial cooperation
There is no industrial cooperation yet. This is partly due to the fundamental character of the research, and partly due to the initial stage the project is in.

International cooperation
In the context of BRICKS we actively collaborate with National ICT Australia, Florida International University, and TU Karlsruhe.

Highlights 2005-2006
Research highlights
As this project is part of the second phase of the BRICKS program (financed through the first open round July 2005), the project is only currently in full swing: the two postdocs started summer 2006. The number of key publications is therefore limited. Three papers accepted at competitive conferences are listed here.

Economic & societal impact
The need for the type of research as performed in this project was presented by the project leader at a mini-symposium on computational geometry and GIS. The audience consisted of academics and GIS specialists from companies and organisations. The event was supported by the Netherlands Geodetic Commission.

Future work 2007-2009
The current research of the consortium has led to new research results that have been "submitted". At the same time we are compiling a list of all results in spatial and spatio-temporal data mining methods, which will be the basis of research at a workshop that will be held on this topic. It will also be the basis of further research done within the project. Other types of geographical analysis (like spatial interpolation, classification, and auto-correlation) will be investigated from the algorithms design angle, which will lead to new, more efficient methods to perform analysis in geographic environments (including anisotropy, scale, and transition). Research on algorithms that determine dependence on data imprecision are investigated in conjunction with research in the NWO open competition project GOGO. Several fundamental results have been obtained, and we will continue to develop new results in this direction. The project leader and the postdocs (Jun Luo and Magdalene Grantson) will work together on the topics of algorithms for spatial data mining and geographical analysis. It is important for the type of research to discuss the best modelling of the problem together, and brainstorm about possible solutions.

For more information, please refer to the publications and posters of this project.